This is actually ok for this function, but initialised to zero anyway if
PURIFY defined.

This does have the impact of masking any *real* unitialised data reads in bn though.

Patch based on approach suggested by Rich Salz.

PR#3415